Accuracy analysis of vegetation cover measurement using digital photography for regular planted woodland

Abstract: Digital photography is now the most widely used method to obtain the vegetation cover in field measurements. It uses a digital camera to shoot the research area straight downward, classify the digital images into plants and soil, and calculate the percentages of the pixel numbers of plants. When measuring the vegetation cover of regular planted woodland with digital photography, the accuracy of the calculated vegetation cover is affected by following factors such as the sampling strategy and non-vertical shooting of digital camera. However, it is difficult to analyze these factors mentioned above quantitatively in situ measurement because there are lots of uncertain errors involved. In this study, we used the computer simulation method to evaluate these factors because in a computer simulated scene, we can strictly control all these uncertainties. We measured a regular planted pyrus orchard to get the structural data in Huailai (40.373°N, 115.723°E), Heibei province. Through statistics of the measured structural data, a regular planted pyrus orchard scene was reconstructed with the plants structural generating software, onyx tree professional which is widely used in forest management, plants reconstruction, and remote sensing of plants. Then the physically based ray-tracing (PBRT) algorithm was used to simulate the illumination conditions of the reconstructed scene and a virtual camera was simulated to get the images taken by a digital camera. The virtual camera settings are just like the digital camera when we measure the vegetation cover in situ. By controlling the shooting conditions strictly within the computer model, we can obtain simulated images under different shooting conditions. The supervised classification method was used to classify the simulated images to obtain the vegetation cover values. Due to the obvious three different samples in the regular planted woodland, this study proposed an area-weighted sampling strategy to calculate the vegetation cover of regular planted woodland. The strategy was then validated in the reconstructed scene using a computer simulation method. It showed that the area-weighted sampling strategy can obtain the vegetation cover of the regular planted woodland accurately, and the relative error is about 7.3% in this simulated pyrus scene. Furthermore, the study analyzed the influence of the non-vertical shooting. We simulated different images taken by the digital camera of these three samples and the viewing zenith angle of digital camera was from -20°~20°. The results showed that the non-vertical shooting had a great impact on the accuracy of the calculated vegetation cover. The relative error of vegetation cover can reach 98.4% for one sample and 27.9% for the calculated vegetation cove of the regular planted pyrus scene. The computer simulation method presented in this paper was demonstrated to be feasible and objective for analyzing the accuracy of digital photography. An area-weighted sampling strategy to calculate the vegetation cover of regular planted woodland and the non-vertical shooting were evaluated in this paper. The results can provide guidance for obtaining the FVC in field measurements using digital photography. Besides, the computer simulation methods came up with in this paper can be used to evaluate the accuracy of other instruments such as LAI2000, or TRAC in following studies.